Introduction: Predictive genomics, integrated with artificial intelligence (AI), represents a pivotal advancement in modern medicine, enabling the anticipation of disease risks, personalized treatment responses, and phenotypic traits through the analysis of genetic data. This review synthesizes recent developments in AI-driven predictive genomics, highlighting its transformative potential in personalized diagnosis and therapy. The objective is to comprehensively examine the latest applications, challenges, and future directions in this interdisciplinary field, drawing from studies published between 2018 and 2025.
Methods: The methodology involved a systematic literature search across databases such as PubMed, Scopus, Web of Science, and specialized genomic repositories like gnomAD and ClinVar. Keywords included combinations of "artificial intelligence," "machine learning," "deep learning," "predictive genomics," "polygenic risk scores," "multi-omics," and "pharmacogenomics." Inclusion criteria focused on peer-reviewed articles in high-impact journals, emphasizing original research, reviews, and meta-analyses with clinical relevance. Over 5,000 initial results were screened, yielding approximately 300 key studies
after title/abstract review and full-text assessment. Quality evaluation utilized tools like PRISMA 2020, AMSTAR 2, and TRIPOD-AI to ensure robustness and minimize bias.
Results: Key findings reveal significant progress in predictive genomics technologies enhanced by AI. Polygenic risk scores (PRS) have evolved to incorporate multi-omics data, improving risk prediction for complex diseases like breast cancer and atrial fibrillation, with accuracy gains of 15-30% in diverse populations. Variant interpretation has advanced through AI models like AlphaMissense, classifying millions of genetic variants with over 90% precision, aiding early detection of rare disorders. Multi-omics integration, using convolutional neural networks (CNNs) and transformers, uncovers intricate biological interactions, enhancing models for disease progression and therapeutic outcomes. Clinical applications demonstrate AI's efficacy in forecasting risks for common ailments—such as cardiovascular events via PRS-proteomics fusion—and in pharmacogenomics, where response to drugs like immunotherapies is predicted with reduced adverse effects. Databases like UK Biobank and All of Us provide vast, diverse datasets, though population biases persist, affecting model generalizability. Challenges include data bias leading to algorithmic discrimination, privacy concerns in genomic data sharing, interpretability issues in black-box models, and limited transferability across demographics and healthcare settings.
Conclusion: In conclusion, AI serves as a cornerstone in elevating personalized medicine by enabling precise, proactive interventions, yet persistent hurdles like bias and explainability demand ongoing refinement. Future research should prioritize multimodal AI, digital twins for patient simulation, explainable frameworks, and regulatory standards like FDA's Predetermined Change Control Plans and TRIPOD-AI to foster equitable, ethical advancements. This synergy promises to revolutionize healthcare, reducing disparities and optimizing outcomes through collaborative, interdisciplinary efforts.
Keywords: Artificial Intelligence ,Predictive Genomics ,Personalized Medicine ,Polygenic Risk Scor